Centrosome-related gene SPHK1 drives bladder cancer progression and therapeutic vulnerability

Highlights

• •The CRSS model effectively stratifies BLCA patients into high- and low-risk groups and predicts bladder cancer survival.
• •CRSS reveals immune infiltration patterns and therapy sensitivity.
• •Machine learning combined with Mendelian randomization identifies key CRGs.
• •SPHK1 drives proliferation and migration of bladder cancer cells.

Abstract

Background

Centrosome-related genes (CRGs) regulate cell division and genomic stability and may influence tumor progression, but their prognostic and functional roles in bladder cancer (BLCA) are not fully defined.

Methods

We curated 698 CRGs and identified 12 prognostic genes to construct a centrosome-related signature score (CRSS). The prognostic value of CRSS was evaluated in TCGA and GSE13507 cohorts. Functional enrichment, immune landscape, and drug sensitivity were analyzed through pathway analysis, ESTIMATE, and IC50 prediction. Machine learning and Mendelian randomization were used to identify the most promising CRGs. Furthermore, we performed single-cell RNA sequencing analysis to uncover the expression patterns of candidate CRGs. Expression and functional validation were performed using RT-qPCR, CCK-8, wound-healing assays, and immunohistochemistry.

Results

CRSS stratified patients into high- and low-risk groups with significant differences in overall survival. High-risk patients exhibited upregulated Aerobic glycolysis (AC), oxidative phosphorylation (OxPhos) and partial epithelial–mesenchymal transition (pEMT) pathways, higher immune/stromal scores, and distinct immune infiltration, including increased M0/M2 macrophages and decreased CD8⁺ T and follicular helper T cells. High-risk tumors showed altered sensitivity to multiple drugs, including AZD8186 and Staurosporine. Integrative analyses identified SPHK1 as a key candidate CRG. SPHK1 expression was associated with tumor grade, stage, poor survival, and the wound-healing molecular subtype. Functional assays showed that SPHK1 knockdown inhibited bladder cancer cell proliferation and migration. Immunohistochemistry further confirmed higher SPHK1 expression in tumors with more aggressive clinicopathological features.

Conclusions

We established a CRG-based model for BLCA and identified SPHK1 as a key gene associated with tumor progression and clinical outcomes.

Graphical abstract

Introduction

BLCA is a globally widespread malignancy, particularly prevalent among men [1,2]. It exhibits a high degree of heterogeneity, with diverse pathological characteristics and molecular subtypes [3]. Despite the emergence of an increasing number of urinary biomarker tests, improved imaging techniques, and blood-based assays based on biomarkers in the field of BLCA, the prognoses for BLCA patients remain unsatisfactory [2,4,5]. Research indicates that bladder cancer patients are often not easily detected in the early stages, and those with advanced or high-risk bladder cancer tend to have worse prognoses and treatment outcomes [6]. Existing markers and diagnostic tools face challenges in accurately forecasting bladder cancer prognosis and treatment response [7,8]. Hence, there is a pressing demand to create efficient and new biomarkers and predictive models to improve survival rates for bladder cancer patients.

A pair of centrioles and the surrounding pericentriolar material constitutes a centrosome [[9], [10], [11]]. A significant hallmark of cancer is the abnormal amplification of the centrosome. Highly active centrosomes in cancer cells can result in irregular cell division, chromosomal instability, and further promotion of cancer cell proliferation [12,13]. Furthermore, there is an association between the centrosome and immune responses and tumor resistance, which directly affects the outcomes of cancer treatment. Research showed that the loss of the BRCA1 cancer gene can lead to centrosomal dysfunction, thereby promoting carcinogenesis in specific tissues [14]. Some studies have demonstrated that centrosome abnormalities in bladder cancer cells are closely associated with chromosomal instability, particularly in the context of p53 mutations. Such alterations may drive centrosome amplification, thereby contributing to the progression of BLCA [15]. The centrosome has long been investigated as a potential target for novel therapeutic strategies. However, there is currently no research on prognostic models related to centrosomes in bladder cancer. Therefore, studying genes related to the centrosome in bladder cancer provide a research direction for the development of more effective bladder cancer early diagnostic methods in the future.

In this study, we aimed to systematically explore the prognostic value of CRGs in BLCA and construct a centrosome-related signature for risk stratification. We further integrated multi-omics analyses, including machine learning, Mendelian randomization, and single-cell RNA sequencing, to identify key candidate genes associated with tumor progression. Functional experiments were performed to validate the biological roles of representative genes in BLCA, with a focus on their potential clinical implications for prognosis and therapy.

Materials and methods

BLCA dataset procurement and preprocessing

Transcriptomic and clinical data of BLCA patients were obtained from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/) and accessed via the UCSC Xena platform (https://xenabrowser.net/). To enhance the robustness and generalizability of the findings, multiple publicly available datasets were included as external validation cohorts. Specifically, the GEO datasets GSE13507 and GSE48075 were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). The IMvigor210 cohort data were obtained from a publicly available resource (http://research-pub.gene.com/IMvigor210CoreBiologies) and processed using the “IMvigor210CoreBiologies” R package. All datasets used in this study are publicly available, and the corresponding accession numbers and access links have been clearly provided to ensure transparency and reproducibility.

Screening of differentially expressed genes

We collected a list of 726 centrosome-related genes CRGs from the MiCroKiTS database and relevant literatures [16,17]. The expression data of 698 CRGs were obtained from the TCGA-BLCA database. Differential expression analysis was performed using the "Deseq2" R package. The full list of genes is provided in Supplementary Table 1.

Selection of centrosome-related genes

RNA expression across datasets was cross-validated, and prognostic genes were identified via univariate Cox regression and refined by LASSO with cross-validation. Eleven machine learning models-including XGBoost, RF, SVM-FREE, Boruta, GBM, GLM, KNN, PLS, Naive Bayes, stepLDA, and Elastic Net-were trained using caret, interpreted using DALEX, and evaluated by ROC analysis, with variable importance quantified to highlight key contributors.

Functional enrichment analysis

Utilizing the clusterProfiler R package, we conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis.

Assessing the immune microenvironment of BLCA

The ESTIMATE algorithm was employed to compute stromal and immune scores, alongside tumor purity assessment. Additionally, CIBERSORT was utilized to estimate the abundance of tumor-infiltrating immune cells within the tumor immune microenvironment.

Overall Survival (OS) and nomogram construction

The Cox proportional hazards regression model was used to evaluate overall survival (OS). A nomogram was constructed to predict 1-, 3-, and 5-year OS using the rms package in R. External validation of CRGs was performed using the Kaplan–Meier Plotter online database (kmplot.com/analysis/).

Predicting drug responses and immunotherapy sensitivity

The predictive capabilities of risk score-associated chemotherapeutic agents were evaluated using the R package oncoPredict. This involved calculating patients IC50 values for several common chemotherapeutic agents. The Wilcoxon rank test was employed to assess variances in IC50 levels between high-risk and low-risk groups. We used an additional immunotherapy dataset, IMvigor210, which focuses on uroepithelial carcinoma, to predict immunotherapy responses.

Single-cell RNA sequencing (scRNA-seq) analysis

Initially, scRNA-seq data of 9 BLCA tissues and 4 normal adjacent tissues were downloaded from the GEO (GSE222315) database. Single-cell unsupervised clustering was performed using the Seurat package in R, with the read count matrix serving as input. Rigorous quality control criteria were implemented, focusing on the number of detected genes and the proportion of mitochondrial gene count per cell. Cells were filtered out if they exhibited fewer than 200 detected genes or had over 15% mitochondrial gene count. Additionally, genes detected in fewer than 3 cells were excluded from subsequent analysis to reduce unforeseen noise. Integration of multi-sample data and correction of batch effects were performed using Harmony algorithm [18]. Subsequently, following the Seurat-guided tutorial, the analysis proceeded with dimension reduction clustering and differential expression analysis. Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) were conducted using the top 15 principal components.

Mendelian randomization (MR) analysis

SNP-specific effect estimates were obtained using mr_singlesnp, and forest and funnel plots were generated using mr_forest_plot and mr_funnel_plot to visualize SNP-specific effects and assess potential heterogeneity. Sensitivity analysis was conducted using the “leave-one-out” approach to evaluate the influence of individual SNPs on the overall causal estimate, and mr_scatter_plot was used to depict SNP-exposure versus SNP-outcome effects, providing a visual summary of the overall causal relationship. Importantly, MR analysis was not used to directly validate the results of machine learning approaches. Instead, it was performed as an independent complementary strategy to further support the biological relevance and robustness of the key candidate genes identified through machine learning.

Cell culture

We obtained human SV-HUC-1, HEK293T, T24, 5637, J82, and UM-UC-3 cells from the Cell Bank of the Chinese Academy of Sciences. These cells were cultured in DMEM with 10% FBS and 1% antibiotics at 37°C with 5% CO2.

Quantitative real-time PCR (qRT-PCR)

We isolated RNA from cell lines using Trizol (Sigma) and measured its concentration. Then, 1 µg of RNA was reverse transcribed into cDNA using the Rever TraAce qPCR RT kit (Toyobo). Quantitative PCR was conducted with the SYBR Green real-time PCR Master Mix (Toyobo) and analyzed using the StepOne real-time PCR system. Primer sequences are provided in Supplementary Table 2.

Immunohistochemistry analysis

Paraffin-embedded tissue sections (4 μm) were deparaffinized, rehydrated, and subjected to microwave antigen retrieval in citrate buffer (pH 6.0). Endogenous peroxidase and nonspecific binding were blocked, followed by overnight incubation at 4°C with primary antibodies: SPHK1 Rabbit mAb (#12071, Abcam, 1:200) and KI67 Rabbit mAb (#ab16667, Abcam, 1:200). Sections were then incubated with HRP-conjugated goat anti-rabbit secondary antibody (#AB205718, Abcam, 1:1000), developed with DAB, counterstained with hematoxylin, and mounted.

Construction of shRNA and lentiviral infection process

The target sequence for sh1-SPHK1 was 5′-GCAGCTTCCTTGAACCATTAT-3′, and for sh2-SPHK1 was 5′-CCCAAACTACTTCTGGATGGT-3′. The primer sequences for shRNA were synthesized by Shanghai Sangon Biotech. After annealing the oligo fragments in both forward and reverse directions, the annealed products were directly ligated with the PLKO.1-U6-shRNA-Puro vector to form shRNA-SPHK1.

Subsequently, shRNA-SPHK1 along with psPAX2 and pMD2.G were co-transfected into HEK293T cells. After 48 hours, lentivirus was harvested and used to infect T24 cells for 1-2 days. Upon reaching confluency in a 6-well plate, puromycin drug selection was initiated for purification, resulting in stable SPHK1 knockdown T24 cell lines.

Cell proliferation and migration

CCK-8 Assay: Cell proliferation was assessed using a cell proliferation reagent . This reagent can measure cell metabolic activity over a certain period, indirectly evaluating cell proliferation ability; Scratch Assay: A cell scratch device or micropipette tip was used to create a straight line scratch on the surface of cell culture dishes to observe cell migration and scratch closure ability.

Statistical analysis

Statistical analyses were conducted using R version 4.1.1. A significance level of p < 0.05 was chosen to determine statistical significance. Depending on the type of data and research questions, a variety of statistical tests were employed. Specifically, the Wilcoxon test and Fisher's exact test were used when appropriate for non-parametric and categorical data analysis, respectively.

Results

Centrosome-related gene signature predicts prognosis in BLCA

A comprehensive list of CRGs was curated from literature and public databases (Supplementary Table 1), among which 698 genes were confirmed to be expressed in the training cohort. Differential expression analysis identified 195 CRGs between normal and tumor tissues, including 134 upregulated and 61 downregulated genes (Fig. 1A). Twelve prognostic CRGs—MAP1A, MAP1B, FAM110B, SPHK1, DUSP23, XRCC3, USP21, BBC3, GEN1, FAM110A, PSME2, and TRMU—were selected via univariate Cox and LASSO regression and used to construct a centrosome-related signature score (CRSS) (Fig. 1B–D): CRSS = (MAP1A * 0.007) + (MAP1B * 0.082) + (FAM110B * 0.055) + (SPHK1 * 0.077) - (DUSP23 * 0.089) + (XRCC3 * 0.108) + (USP21 * 0.108) - (BBC3 * 0.191) - (GEN1 * 0.134) - (FAM110A * 0.014) - (PSME2 * 0.239) - (TRMU * 0.06).

Fig. 1.

Construction and validation of the centrosome-related prognostic signature in BLCA. (A) Volcano plot of differentially expressed centrosome-related genes between BLCA and normal tissues (p < 0.05, |log2FC| > 1). (B) LASSO Cox regression for selection of prognostic CRGs. (C) Multivariate Cox analysis of 12 candidate genes. (D) Expression levels of 12 model genes in TCGA-BLCA cohort. (E-F) Kaplan–Meier survival analysis of high- and low-risk groups in TCGA and GSE13507 cohorts. (G-H) Distribution of CRSS, survival time, and survival status in TCGA and GSE13507 cohorts. (I-J) Time-dependent ROC curves assessing predictive performance in TCGA and GSE13507 cohorts. Statistical significance: **p < 0.01; ****p < 0.0001.

Based on the optimal cutoff, patients were stratified into high-risk(n = 165) and low-risk (n = 235) groups. Kaplan–Meier analysis demonstrated that high-risk patients had significantly shorter overall survival and a higher incidence of death events. The prognostic performance of the CRSS was further validated in the independent GSE13507 cohort, which showed comparable risk stratification and predictive accuracy. Time-dependent ROC analysis yielded AUCs of 0.66, 0.62, and 0.63 for 1-, 3-, and 5-year overall survival, respectively (Fig. 1B-J). These results demonstrate that the centrosome-related prognostic model robustly captures molecular heterogeneity and provides accurate survival prediction in BLCA.

Development and clinical validation of a centrosome-related prognostic nomogram in BLCA

Univariate and multivariate Cox regression analyses identified age, pathological stage, and the CRSS as independent prognostic factors in bladder cancer (Fig. 2A–B). Based on these variables, we constructed a prognostic nomogram to predict 1-, 3-, and 5-year OS, and calibration curves showed excellent agreement between predicted and observed outcomes (Fig. 2C–D). External validation using the Kaplan–Meier Plotter database confirmed the prognostic significance of CRGs for OS in bladder cancer, consistent with our previous findings (Supplementary Fig. 1A–L).

Fig. 2.

Establishment and evaluation of the nomogram survival model in BLCA. (A) Univariate Cox analysis of clinicopathological factors and risk score in the TCGA-BLCA cohort. (B) Multivariate Cox analysis of clinicopathological factors and risk score. (C) Construction of a nomogram predicting 1-, 3-, and 5-year overall survival. (D) Calibration curves evaluating the accuracy of the nomogram in TCGA-BLCA cohort. (E–H) Associations between risk score and age, tumor grade, gender, and papillary features. (I) Combined ROC curves showing predictive performance of the centrosome-related model at 1-, 3-, and 5-year time points.

Further analysis revealed that patients aged >65 years or with poorly differentiated tumors exhibited higher risk scores, reflecting their association with aggressive tumor phenotypes and unfavorable prognosis, whereas female patients and those with papillary histology generally had lower risk scores. The integrated risk model achieved AUCs of 0.733, 0.743, and 0.739 for 1-, 3-, and 5-year OS, respectively (Fig. 2E–I), indicating robust predictive performance across multiple time points.

Enrichment analyses of high/low-risk group patients

To comprehensively explore the potential mechanisms underlying CRSS in different risk groups, we conducted GO and KEGG enrichment analyses. These analyses revealed that in the high-risk group, there was an upregulation of processes related to response to xenobiotic stimulus, Metabolism of xenobiotics by cytochrome P450, and the PPAR signaling pathway. In contrast, the low-risk group exhibited enrichments in functions related to supporting cell survival, intercellular communication, and the physiological functions of tissues (Supplementary Fig. 2A-D).

To further elucidate the molecular and immune characteristics between risk groups, we assessed the activity of 16 cancer-related pathways. High-risk patients showed elevated activity in the AC, Oxphos and pEMT pathways, suggesting enhanced metastatic potential. Interferon signaling and several immune-related pathways were also more active in the high-risk group. Consistently, immune profiling revealed that high-risk patients had increased levels of APC co-stimulation, B cells, macrophages, mast cells, and neutrophils (Fig. 3A-B).

Fig. 3.

Distinct immune and metabolic landscapes between high-risk and low-risk BLCA patients. (A) Differences in immune-related pathway activities between groups. (B) Differences in metabolic pathway activities between groups. (C–F) Comparison of ESTIMATE score, immune score, stromal score, and tumor purity between groups. (G) Relative proportions of 22 immune cell types estimated by CIBERSORT. Statistical significance: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, not significant.

Given the strong association between the centrosome-related model and immune regulation, we next examined the tumor immune microenvironment. The ESTIMATE algorithm revealed higher immune and stromal scores but lower tumor purity in high-risk patients (Fig. 3C-F). Moreover, high-risk tumors exhibited enrichment of M0 and M2 macrophages and depletion of CD8⁺ T cells, follicular helper T cells, and plasma B cells (Fig. 3G). Together, these findings highlight distinct molecular and immune landscapes in high-risk bladder cancer, offering insights into tumor progression and potential therapeutic targets.

Distinct responses of immunotherapy and diverse anti-cancer drugs among high/low risk patients

To further explore the clinical relevance of these immune features, we assessed their association with immunotherapy response. Patients receiving immunotherapy showed lower risk scores and improved survival outcomes (Fig. 4A–B). The risk score showed positive correlations with IL6, TNFSF4, and ENTPD1, but negative correlations with PD-L1, CTLA-4, VEGFA, and TNFRSF14. These patterns indicate that low-risk patients may be more responsive to PD-1/PD-L1 or CTLA-4 blockade, whereas high-risk patients may benefit from targeting IL6 signaling, OX40/OX40L, or the CD39/CD73 axis (Fig. 4C). We further evaluated drug sensitivity and found that high-risk patients exhibited lower IC50 values for AZD8186, AZD8055, Staurosporine, KU-55933, RO-3306, BMS-754807, and NU7441. Collectively, these findings highlight distinct immunological and therapeutic vulnerabilities across risk groups and provide a framework for tailored treatment strategies in bladder cancer (Fig. 4D).

Fig. 4.

Associations between the centrosome-related risk score, immunotherapy response, immune modulation, and drug sensitivity in BLCA. (A) Comparison of the centrosome-related risk score between patients who received immunotherapy and those who did not in the IMvigor210 cohort. (B) Kaplan–Meier survival curves showing overall survival differences according to the centrosome-related risk score in the IMvigor210 cohort. (C) Correlation between immunomodulator expression and the centrosome-related risk score in the TCGA BLCA cohort. (D) Associations among drugs, the risk score, and model genes. Statistical significance: *p < 0.05; **p < 0.01; ***p < 0.001.

The expression patterns of candidate CRGs revealed by scRNA-seq analysis

We integrated nine published bladder cancer and four adjacent cancer single-cell datasets and visualized the high-resolution transcriptional landscape of bladder cancer using Uniform Manifold Approximation and Projection (UMAP) (Fig. 5A). Subsequently, we annotated epithelial cells, endothelial cells, stromal cells, cancer-associated fibroblasts (CAFs), myeloid cells, T cells, B cells, and plasma cells based on classical cell markers (Fig. 5A-B). Furthermore, we aimed to decipher which cell subpopulations express CRGs in our risk model. As shown in the UMAP plot, MAP1A is predominantly expressed in Pericytes (Fig. 5C). Moreover, PSME2 is universally expressed across subpopulations, with a particular concentration in T cells (Fig. 5D). Additionally, DUSP23 is weakly expressed only in Endothelial cells (Fig. 5E).

Fig. 5.

Bladder cancer patients' different cellular compositions at single-cell resolution. (A) Cell clustering into 10 types using UMAP dimensionality reduction algorithm, with each color representing a distinct cell population. (B) Dot plots showing the expression of the top 3 marker genes for each cluster. (C-E) Expression of MAP1B, PSME2, and DUSP23 genes in each cluster. (F-G) Distribution of cells from bladder cancer single-cell dataset GSE222315 in bladder cancer adjacent tissue and cancer tissue. (H) Expression of PSME2 in epithelial cells.

We also compared the distribution of different cell types between adjacent bladder cancer and bladder cancer tissues and observed significant differences in the distribution of epithelial cells between the two (Fig. 5F-G). Based on this, we further compared the expression of prognostic model genes in epithelial cells, revealing that PSME2 expression in epithelial cells of adjacent tissues is higher than that in cancer tissues (Fig. 5H). In summary, we validated the expression landscape of centrosome-related genes at the single-cell level and demonstrated the sensitive predictive ability of PSME2 in BLCA at the single-cell level.

Machine learning–mendelian randomization analysis

Given the molecular heterogeneity of bladder cancer and the complex regulation of centrosome-associated pathways, we next aimed to identify the CRG with the greatest prognostic and functional significance. To this end, we combined machine learning with Mendelian randomization analyses. Twelve prognostic CRGs were systematically evaluated across eleven machine learning algorithms—including BORUTA, RF, XGBoost, SVM, decision tree, Elastic Net, GBM, GLM, KNN, Naive Bayes, and stepLDA—to assess their predictive contributions (Fig. 6A–G). Mendelian randomization provided complementary causal evidence linking candidate genes to disease traits (Fig. 7A–D). Convergent results from these independent approaches consistently highlighted SPHK1 as the most robust CRG, underscoring its potential role as a driver of tumor progression and a promising target for immunotherapy in BLCA (Fig. 8A–C).

Fig. 6.

Machine learning evaluation of prognostic CRGs. (A–G) The predictive contributions of 12 prognostic CRGs were assessed across eleven machine learning algorithms. Feature importance and model performance highlight SPHK1 as a top predictive gene.

Fig. 7.

Mendelian randomization analysis of centrosome-related genes in BLCA. (A) Single-SNP effect estimates showing associations between candidate CRGs and BLCA risk. (B) Forest plots illustrating causal effects of individual SNPs on disease outcomes. (C) Funnel plots assessing heterogeneity among instrumental variables. (D) Sensitivity analyses validating the robustness of causal inferences.

Fig. 8.

Clinical relevance of SPHK1 in BLCA. (A–C) Venn diagram from integrated machine learning and Mendelian randomization analyses identifying SPHK1 as the most robust prognostic centrosome-related gene. (D–I) SPHK1 expression across tumor grades and stages, and its association with overall survival in TCGA, GSE13507, and IMvigor210 cohorts, highlighting its prognostic significance. (J) Association between SPHK1 expression and therapeutic response. (K) Correlation of SPHK1 expression with clinical and molecular subtypes in TCGA. (L) Correlation of SPHK1 expression with immune subtypes in TCGA.

SPHK1 drives BLCA progression and subtype features

To further assess the clinical relevance of SPHK1 in BLCA, patient datasets were analyzed. SPHK1 was highly expressed in BLCA tissues in the TCGA cohort, with expression increasing alongside tumor grade and stage, but no significant differences were observed in M stage. High SPHK1 expression was significantly associated with poor treatment response and increased mortality, indicating worse prognosis. These findings were further validated in the external GSE13507 and IMvigor210 cohorts (Fig. 8D–I). Moreover, SPHK1 expression distinguished molecular subtypes, with high expression predominantly enriched in the C2 (wound-healing) subtype (Fig. 8J–L). Collectively, these results suggest that SPHK1 may play a critical role in BLCA progression and prognosis.

Validation of 12 centrosome prognostic model genes through in vitro experiments

To further evaluate the expression patterns of the 12 CRGs, we conducted rigorous RT-qPCR validation across a range of cell samples, including normal bladder cells (SV-HUC-1) and four distinct BLCA cell lines (T24, 5637, J82, UM-UC-3). Compared to SV-HUC-1 cells, our findings revealed that the 5637 bladder cancer cell line exhibited notably elevated mRNA levels for BBC3, DUSP23, FAM110A, GEN1, and PSME2. In the case of the J82 bladder cancer cell line, higher mRNA levels were observed for BBC3, DUSP23, PSME2, TRMU, USP21, and XRCC3. The T24 bladder cancer cell line displayed elevated mRNA levels for BBC3, USP21, and XRCC3. Meanwhile, the UM-UC-3 bladder cancer cell line showcased increased mRNA levels for GEN1. Remarkably, FAM110B displayed a converse trend in these four cell lines. Notably, there were no discernible differences in the expression levels of MAP1A, MAP1B, and SPHK1 across the diverse cell types examined (Fig. 9A-L). These findings corroborate well with our analysis results in the TCGA cohort overall.

Fig. 9.

RT-qPCR analysis of 12 centrosome-related genes in bladder cancer cell lines (A–L). Statistical significance: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Functional validation of SPHK1 in bladder cancer progression

To further investigate the role of SPHK1 in bladder cancer cell proliferation and migration, we established SPHK1-knockdown T24 cells (Fig. 10A). CCK-8 and wound-healing assays showed that SPHK1 knockdown significantly inhibited cell proliferation and reduced migration capacity (Fig. 10A–D). Immunohistochemistry of a self-constructed cohort (56 BLCA tumor and adjacent tissues) and the HPA database (ID: HPA028761) consistently demonstrated that SPHK1 protein levels were markedly higher in tumor tissues than in adjacent normal tissues. High SPHK1 expression was associated with increased Ki67 levels, larger tumor size, more frequent lymph node metastasis, advanced stage, and poorer prognosis, suggesting that SPHK1 may serve as a potential therapeutic target in BLCA (Fig. 10E-M). These results not only validate the bioinformatics predictions but also highlight the functional role of SPHK1 in driving bladder cancer progression.

Fig. 10.

Functional validation of SPHK1 in BLCA cells and tissues. (A) Knockdown efficiency of SPHK1 in T24 cells measured by qPCR. (B) Effect of SPHK1 knockdown on cell proliferation assessed by CCK-8 assay. (C–D) Impact of SPHK1 knockdown on cell invasion and migration evaluated by scratch assay. (E–M) Immunohistochemistry in a self-constructed BLCA cohort (n = 56) and HPA database (ID: HPA028761) revealed higher SPHK1 protein levels in tumors compared with adjacent tissues. High SPHK1 expression was associated with increased Ki67, larger tumor size, lymph node metastasis, advanced stage, and poorer prognosis. Statistical significance: *p < 0.05; **p < 0.01; ***p < 0.001.

These comprehensive findings not only enhanced the credibility of the earlier bioinformatics analysis but also provided substantial validation of the clinical relevance of the predictive model.

Discussion

BLCA is a common malignancy in the genitourinary system [19]. Current treatment strategies include chemotherapy, targeted therapy, and immunotherapy; however, prognosis remains poor for many patients [20,21]. Traditional clinical and pathological methods are insufficient for accurately predicting disease progression and survival in BLCA patients [22,23]. Therefore, the identification of effective biomarkers and therapeutic targets is essential for improving patient survival rates.

The centrosome plays a crucial role in mitosis [24]. Centrosomes must self-replicate before cell division to ensure accurate chromosome segregation in each daughter cell [25]. However, abnormalities in centrosome replication or function can lead to issues during mitosis. An excess of centrosomes can lead to spindle abnormalities, increasing the risk of errors in chromosome segregation [[26], [27], [28]]. Notably, the development of bladder cancer is closely linked to cell cycle dysregulation [29]. Centrosomal abnormalities can induce chromosomal instability and aneuploidy, which are key features of cancer and may promote bladder cancer progression [30]. Understanding the role of centrosomal anomalies in BLCA pathogenesis is critical for the development of targeted therapies aimed at restoring centrosome homeostasis or exploiting centrosome-related vulnerabilities in cancer cells.

In this study, we comprehensively characterized CRGs in BLCA through integrated bioinformatics, machine learning, and experimental validation. Among 698 expressed CRGs, twelve were identified as prognostically relevant. These genes are functionally implicated in centrosome regulation and may drive BLCA progression. Specifically, MAP1A and MAP1B belong to the microtubule-associated protein family and interact with both microtubules and actin filaments [31,32], suggesting roles in centrosome stabilization and mitotic spindle organization. SPHK1 and TRMU are involved in lipid and energy metabolism, which may impact processes related to cell division and centrosome function [[33], [34], [35]]. Loss of DUSP23 impairs centrosome duplication and induces spindle pole defects during mitosis [36]. USP21 modulates the FOXM1 transcriptional network, thereby affecting cell cycle progression and proliferation [37]. PSME2, as a component of the protein degradation and antigen presentation system, may have an impact on cell mitosis and centrosome function [38]. GEN1 contributes to genome maintenance, and its dysregulation may lead to chromosomal instability [39]. XRCC3 plays a role in homologous recombination-mediated DNA repair [40]. FAM110 proteins localizes to the centrosome and accumulates at the microtubule organizing center during interphase [41]. Collectively, these findings highlight the functional relevance of CRGs in maintaining centrosome integrity and genomic stability, providing mechanistic insights and potential therapeutic targets for BLCA.

Machine learning and Mendelian randomization analyses consistently identified SPHK1 as a potential driver gene in bladder cancer. Increasing evidence suggests that SPHK1 not only promotes tumor cell proliferation and migration but also plays a critical role in remodeling the tumor immune microenvironment and enabling immune evasion. Notably, in colorectal liver metastasis, targeting SPHK1 in tumor-associated macrophages reprogrammed the immunosuppressive microenvironment and enhanced the efficacy of anti–PD-1 immunotherapy [42]. Similarly, in head and neck squamous cell carcinoma, SPHK1 promotes immune escape via the MMP1–PD-L1 axis, suppressing anti-tumor immunity and supporting tumor progression [43]. In bladder cancer, we found that high SPHK1 expression was associated with an immunosuppressive tumor microenvironment, characterized by increased M2 macrophage infiltration, reduced CD8⁺ T cell abundance, and enrichment in the wound-healing molecular subtype. Functional assays confirmed that SPHK1 promotes bladder cancer cell proliferation and migration, reinforcing its oncogenic role. These results suggest that SPHK1 contributes to establishing an immunosuppressive niche in bladder cancer through mechanisms analogous to those reported in other tumor types.

It cannot be denied that our study has some limitations. The actual value and clinical applicability of the centrosome-related prognostic model need further validation in larger prospective trials to unveil their inherent connections and complex regulatory mechanisms.

In conclusion, this study has not only improved our understanding of the centrality of the centrosome in BLCA but has also provided a robust prognostic model with clinical value. The centrosome-related model can aid in assessing patient risk and personalizing treatment approaches, ultimately enhancing outcomes in bladder cancer management.

Informed consent statement

The need for informed consent was waived by the Ethics Committee, as the study used publicly available datasets and de-identified clinical samples with no direct involvement of human subjects.

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Shanghai TCM-Integrated Hospital, Shanghai University of Traditional Chinese Medicine (Approval Nos. 2024-043-1 and 2025-042-1).

Clinical trial registration

Not applicable.

Funding statement

This work was supported by the Shanghai Municipal Health Commission Clinical Research Special Project [20254Y0028], the 2024 Science and Technology Development Program of Shanghai University of Traditional Chinese Medicine (Natural Science Category) [24KFL083], the Young Investigator Program of Shanghai Integrated Traditional Chinese and Western Medicine Hospital [2024KJCY008], and the Hongkou District Health Commission Traditional Chinese Medicine Research Project [HKZYY-2025-26].

CRediT authorship contribution statement

Chengyuan Dong: Writing – original draft, Visualization, Validation, Funding acquisition, Data curation, Conceptualization. Zhou Fang: Writing – review & editing, Validation, Methodology, Formal analysis, Data curation. Yiling Han: Writing – review & editing, Formal analysis, Data curation. Yuan Yin: Writing – review & editing, Supervision, Resources, Project administration. Yongbing Cao: Writing – review & editing, Supervision, Resources, Project administration, Funding acquisition. Qun Lu: Writing – review & editing, Visualization, Supervision, Resources, Project administration, Funding acquisition.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The public datasets used in this study include TCGA, GEO, and IMvigor210, with detailed descriptions provided in the Methods section. The in-house cohort data generated in this study are available from the corresponding author upon reasonable request.